AI-Enhanced Problem-Based Learning Framework: Integrating ChatGPT as Adaptive Scaffolding to Improve Critical Thinking and Personalized Learning

Synthesis: La Sunra, Amaliah, & Radhiyani (2026) examine an AI-Enhanced Problem-Based Learning (AIPBL) framework that integrates ChatGPT as adaptive scaffolding to improve critical thinking and support personalized learning. Using an explanatory sequential mixed-methods design with 120 eighth-grade students across four junior high schools in Makassar, they find a large, significant improvement in critical thinking — from 60.3 (SD 9.7) to 73.5 (SD 10.4), t(119) = 9.64, p < .001, d = 1.00 — with notable gains in analysis, inference, and bias detection. Qualitative analysis shows ChatGPT supported idea exploration, strengthened verification habits, and enhanced reflection and metacognitive monitoring, without producing passive dependence.

Key Findings

• Large critical thinking gains. Overall critical thinking scores rose from 60.3 (SD 9.7) to 73.5 (SD 10.4), t(119) = 9.64, p < .001, d = 1.00 — a large effect — with notable gains in analysis, inference, and bias detection.
• Four qualitative patterns. ChatGPT supported idea exploration, strengthened students' verification habits, enhanced reflection and metacognitive monitoring, and — crucially — did not lead to passive dependence when embedded in structured PBL.
• ChatGPT as adaptive scaffolding. The educational value of ChatGPT depends on how it is used: it functions as a learning partner when embedded in structured PBL and guided by teachers, rather than as a standalone answer machine.
• Personalized learning support. The framework used ChatGPT to tailor scaffolding to individual student needs, supporting personalized learning within a problem-based format.
• Agency preserved. The study offers practical evidence that ChatGPT can be integrated in secondary classrooms without reducing student agency.

Study Design & Method

An explanatory sequential mixed-methods study with 120 eighth-grade students from four junior high schools in Ujung Pandang District, Makassar. The quantitative phase used a one-group pre-test–post-test design to measure whether students' critical thinking changed after the AIPBL intervention (with ChatGPT as adaptive scaffolding); the qualitative phase examined how students used ChatGPT. This design captured both the extent of improvement and the process behind it.

What this means for practice

• Instructors. Embed ChatGPT inside a structured, teacher-guided problem-based sequence instead of offering it as a general answer machine: that structure is what produced the gains (60.3 to 73.5, d = 1.00) without passive dependence, because design and facilitation decide whether the tool acts as a learning partner or a shortcut.
• Instructors. Build verification routines into tasks explicitly, requiring students to check claims, ask for exceptions, and name the evidence behind a statement — students' own questioning shifted this way over the eight sessions.
• Designers. Add reflective checkpoints and prompt-quality work to the cycle, since students who monitored prompt clarity revised both their questions and their reasoning.
• Designers. Position GenAI output as provisional information within the activity, and design the scaffolding so that the question — not only the answer — becomes a visible learning object.
• Instructors. Keep school-level AI guidance aligned with classroom practice, as the authors recommend for schools beginning to address GenAI.

Limitations

• The quantitative phase used a one-group pre-test–post-test design over eight weeks with no control group, so the authors state the improvement cannot be fully isolated from other external influences.
• The 120 eighth-grade students came from four junior high schools in one district; the schools were selected for comparable implementation settings rather than to represent Indonesian junior high schools, and the authors warn against automatic generalization to schools with different resources or teacher readiness.
• The qualitative phase rests on a purposive subset of 12 students and four teachers across the four classes.
• Outcomes were limited to critical thinking and classroom interaction: long-term retention, transfer to other subjects, and changes in writing quality were not measured.

Citation

La Sunra, S., Amaliah, S., & Radhiyani, F. (2026). AI-enhanced problem-based learning framework: integrating ChatGPT as adaptive scaffolding to improve critical thinking and personalized learning. JEELS, 13(2).